compact: set unready if halted

[JensErat]

If comapctor is halted, it should not report ready state any more.

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Changes

Verification

[bwplotka]

Thanks! But does this mean that Kubernetes will restart it, right?

So this means this change is the same as the setting --no-halt-on-error flag (: No? We actually want still access UI's and everything is working just fine on halt, just compaction is halted, so from HTTP and UI perspective it's ready. What do you think? 🤔

[JensErat]

Seems I was somehow to fast in sending the PR through the command line (actually wanted to add some more notes on exactly that this might be subject of discussion regarding whether this is wanted), but seems I actually sent the first copy instead of cancelling... This PR also still misses changelog and everything because I wanted to be sure this gets agreement first.

I guess in this case you probably should remove the health check? In this case Kubernetes will not restart it any more. I know I can also set the hidden debug flag (actually, that's what I'm deploying right now because of #2805), but saying "compactor is ready" although it actually failed feels slightly wrong.

[GiedriusS]

Thanks! But does this mean that Kubernetes will restart it, right?

So this means this change is the same as the setting --no-halt-on-error flag (: No? We actually want still access UI's and everything is working just fine on halt, just compaction is halted, so from HTTP and UI perspective it's ready. What do you think? thinking

I concur with @bwplotka here. That flag should suffice. Plus, the whole idea is that when Thanos Compactor is halted like this, it should still remain up so that you could go and troubleshoot what had happened.

[bwplotka]

Yes. Essentially if we mark it as unready we could equally just not halt it and crash (: Same effect.

[kakkoyun]

On the other note, bingo requires at least go 1.14